本项目利用SentenceTransformer对字典词条进行向量化，创建Faiss向量索引，实现错别字、同音字、形似字、漏字等情况的搜索词召回。

• Python 3.11+

• 单氨胖光酸 http://localhost:8080/search?word=单氨胖光酸&top=1&pinyin=1

  {
    "code": 1,
    "message": "success",
    "result": [
      {
        "index": "WORD",
        "code": "15273",
        "word": "甘草酸单铵半胱氨酸氯化钠注射液",
        "score": 5,
        "distance": 0.375848472118378
      }
    ],
    "micro": 110706
  }

• 阿代那非 http://localhost:8080/search?word=阿代那非&top=1&pinyin=1

  {
    "code": 1,
    "message": "success",
    "result": [
      {
        "index": "WORD",
        "code": "24317",
        "word": "阿伐那非片",
        "score": 7,
        "distance": 0.395916491746902
      }
    ],
    "micro": 154205
  }

• 霜瓜唐安http://localhost:8080/search?word=霜瓜唐安&top=1&pinyin=1

  {
    "code": 1,
    "message": "success",
    "result": [
      {
        "index": "PINYIN",
        "code": "4598",
        "word": "双瓜糖安胶囊",
        "score": 4,
        "distance": 0.0730657055974007
      }
    ],
    "micro": 129010
  }

• dict/dict_words.csv：字典文件，每行一个词条
• 包含两个字符串字段，分别为: 词条代码、词条内容
• 可通过http(post): /put 接口上传字典文件

  curl -X POST -F "[email protected]" http://localhost:8080/put

• 查看帮助
  • 源码运行

    python main.py -h

  • 可执行文件运行

    cd /path/to/vector-search
    ./vector-search -h

• 创建索引
  • 源码运行

    python main.py index -worker=4 -batch=1000 -min=2 -max=4

  • 可执行文件运行

    cd /path/to/vector-search
    ./vector-search index -worker=4 -batch=1000 -min=2 -max=4

• 启动服务

  • 源码运行

    python main.py server -port=8080 -log-level=info

  • 可执行文件运行

    cd /path/to/vector-search
    ./vector-search server -port=8080 -log-level=info

• 访问帮助页面

  • http://localhost:8080/

• 从 release page 下载Windows打包程序

• 创建虚拟环境

  python -m venv .venv
  source venv/bin/activate

• 安装依赖

  pip install -r requirements.txt

• 下载预训练模型

  from sentence_transformers import SentenceTransformer
  
  model = SentenceTransformer('sentence-transformers/distiluse-base-multilingual-cased-v1')
  model.save('model/distiluse-base-multilingual-cased-v1')

• 打包程序

  pyinstaller ./vector-search.spec

• 本例使用CPU，有GUP的情况下，可使用GPU加速，请修改faiss-cpu为faiss-gpu